import pickle
import numpy as np
import pandas as pd
import os
import sys
from sklearn.tree import DecisionTreeClassifier
from sklearn.svm import LinearSVC
from sklearn.metrics import accuracy_score
from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import train_test_split
from sklearn.linear_model import RidgeClassifierCV
from sklearn.linear_model import LogisticRegressionCV
# I have pulled the initial win rates of champions in patch 11.24 in ARAM, but I will be fine-tuning these through the thousands of games analyzed. I will then convert the
# raw rating of each champion to a number between 0-100 representing the champion's "strength" in ARAM games (not necessarily winrate) through the following logistic function.
def score(x,a=1):
return (1/(1+np.exp(-a*(x))))*100
# Get dataframe
df = pickle.load(open(os.path.join(sys.path[0], "df.p"), "rb"))
df.set_index('MatchID',inplace=True)
# Split into features vs target
X = df.drop("BlueWin",axis=1)
y = df.BlueWin.astype('int')
# Get train and test sets
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=10)
# Set random seed
np.random.seed()
# Logistic Regression CV
lr = LogisticRegressionCV(cv=5, random_state=0).fit(X_train, y_train)
# Ridge Regression CV
rc = RidgeClassifierCV(cv=5).fit(X_train, y_train)
# Parameters for Max Depth for SVM
parameters_svm = {'C':np.logspace(-3,3,5)}
svm = GridSearchCV(LinearSVC(), parameters_svm, n_jobs=-1).fit(X_train,y_train).best_estimator_
# Parameters for Max Depth for Decision Tree
parameters_dt = {'max_depth':range(20,100,5)}
dt = GridSearchCV(DecisionTreeClassifier(), parameters_dt, n_jobs=-1).fit(X_train,y_train).best_estimator_
classifiers = [lr,rc,svm,dt]
algos = ["Logistic Regression","Ridge Regression","SVM","Decision Tree"]
for i in range(4):
print("{}: train accuracy was: {:.4f} and test accuracy was {:.4f}".format(algos[i],accuracy_score(y_train,classifiers[i].predict(X_train)),accuracy_score(y_test,classifiers[i].predict(X_test))))
# pickle fitted logistic regression object
pickle.dump(lr, open("/home/cs/cmsc320/arman-b.github.io/logistic_regression.p", "wb"))
# Create champion ranking list from the coefficients
rankings = pd.DataFrame({'Champion':X_train.columns,'Score':score(lr.coef_.flatten(),7)}).sort_values('Score',ascending=False).reset_index(drop=True)
rankings.Score = rankings.Score.round(5)
rankings.reset_index(inplace=True)
rankings.columns = ["Rank","Champion","Score"]
rankings.Rank += 1 # Rank is 1-indexed
pickle.dump(rankings ,open( "/home/cs/cmsc320/arman-b.github.io/scores.p", "wb" ))
rankings
